Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!watmath!clyde!bonnie!akgua!whuxlm!harpo!decvax!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.micro.68k
Subject: Re: Re: FLAME!!! Re: EA orthogonality
Message-ID: <2252@sun.uucp>
Date: Sat, 1-Jun-85 01:16:22 EDT
Article-I.D.: sun.2252
Posted: Sat Jun  1 01:16:22 1985
Date-Received: Wed, 5-Jun-85 01:08:13 EDT
References: <5609@utzoo.UUCP> <75600002@ima.UUCP> <387@gumby.UUCP>
Organization: Sun Microsystems, Inc.
Lines: 39

> The problem is that C is a programming language written with a particular
> machine storage model in mind, and it ports poorly to other architectures.
> Modula-2, Pascal, Ada, all are languages that port quite well to the 8086
> family, and produce efficient, readable code without any sort of trickery
> required of the programmer.

Umm... Pascal has pointers, "new", and "dispose", just like C has pointers
and its library has "malloc" and "free".  How would you write a program
that, say, manipulated large trees requiring >64KB worth of node storage in
Pascal?  Probably similarly to how you'd write it in C.  Now imagine that
program dealing with two nodes at the same time.  Well, you load one pointer
into the DS register and one of the general-purpose registers, and the other
one into the ES register and one of the general purpose registers.  You have
to kludge a bit with the "use the ES register" prefix, but the advance
information data sheet I have in front of me doesn't list any times for the
segment-selection prefix so I assume it takes no time.

Now imagine that program dealing with three nodes at the same time - say
it's an expression tree, and it's evaluating an addition by adding the LHS
to the RHS and storing the result in the parent node.  Well, you load one
pointer into the DS register and one of the general purpose registers, the
second one into the ES register and one of the general purpose registers,
and the third one into the EES register and one of the general purpose
registers... Oops.  There *is* no EES register.  Oh well, shuffle shuffle
shuffle....  (If you can get good code for this one, try something where you
have to use each pointer more than once.)

The same data sheet says a direct intersegment call takes 19 more clocks
than a direct within segment call in protected mode, and even in real
address mode it takes 6 more clocks.  Load pointer to DS/ES instructions
take 21 clocks in protected mode and 7 in real address mode.  Using those
segment registers dynamically rather than statically is not cheap.

Do you have evidence that compilers for the other languages you mention can
solve problems like this much better than compilers of equivalent quality
for C?  If so, can you show that the difference is due to some
characteristic of the languages in question?

	Guy Harris